Extreme Long Video Comprehension on LVBench
51.4AccuracyVideoExplorer
Evaluation results
| Method | Links | |
|---|---|---|
| VideoExplorerModel Type=Agentic LLMs, Base Model=Qwen2.5-7B-tuning + Qwen2.5-VL-32B2026.01 | 51.4 | |
| GPT-4o#Frames=384, Model Type=Closed-source Models2026.01 | 48.9 | |
| VideoThinker#Frames=64, Model Type=Agentic LLMs, Base Model=Qwen2.5VL-7B2026.01 | 48.9 | |
| Qwen2.5VL-72B#Frames=128, Model Type=Open-source Models2026.01 | 47.4 | |
| Seed1.5VL-pro#Frames=32, Model Type=Closed-source Models2026.01 | 46.1 | |
| Language AgentModel Type=Agentic LLMs, Base Model=Qwen-235B w/ our tools2026.01 | 45.4 | |
| Qwen2.5VL-7B#Frames=64, Model Type=Open-source Models2026.01 | 38.3 | |
| Gemini-1.5-pro#Frames=256, Model Type=Closed-source Models2026.01 | 33.1 |